On algebraic properties of low rank approximations of Prony systems (Q2007942): Difference between revisions

If \(\delta\) denotes the Dirac distribution, the authors reconstruct so-called spike train signals \[ F(x) = \sum_{j=1}^d a_j\,\delta(x-x_j)\,, \quad x \in \mathbb R, \] from given noisy measurements \(\nu_k\) of moments \[ m_k(F) = \int x^k\,F(x)\,dx =\sum_{j=1}^d a_j\,x_j^k\,, \quad k=0,\ldots,N, \] where the nodes \(x_1 < x_2 < \ldots < x_d\), \(d>1\), and the amplitudes \(a_j \in \mathbb R \setminus \{0\}\) of the signal are unknown and where \(|\nu_k - m_k(F)| \le \varepsilon\). The goal is to solve the Prony system \[ \sum_{j=1}^d a_j\,x_j^k = \nu_k\,, \quad k=0,\ldots,N, \] so as to minimize the worst case reconstruction error. The authors describe the geometry of the set of all spike train signals \(F\) with \(d\) nodes such that there exists a signal \({\tilde F}(x) ={\tilde a}_1 \delta(x- {\tilde x}_1)\) with \(m_k(\tilde F) = m_k(F)\), \(k =0,\,1,\,2\). Using low-rank approximation, they present lower bounds on the moment differences between a spike train signal \(F\) with \(d\) nodes and a spike train signal \(\tilde F\) with at most \(d-1\) nodes. Signals with near-colliding nodes are considered, too.